Scholarlink

<p>Accurate identification of defects in Jet Engine Turbine and compressor blade surface plays a vital role in ensuring engine efficiency, safety and secure life. Traditional detection techniques were manual and consumed more time, error prone and need of specialized experts. To address these challenges, this work proposes an automated defect detection system using Swin Transformer Based Deep learning model. High resolution blade surface images were captured, pre-processed and augmented to improve robustness with varying surface condition. The novelty of the proposed work is that it adapts the Swin Transformer architecture to the specific challenges of turbine and compressor blade surfaces and it can detect micro-scale defects with high fidelity, by detecting both local and global features. Experimental results demonstrated that the proposed Swin transformer model produces high detection performance compared to the conventional CNN model with an accuracy of 98.4%, precision of 97.9%, recall of 98.7%, F1-score of 98.3% and mean Average Precision (mAP) of 97.6% on a dataset consisting of 172 high-resolution turbine and compressor blade images. The performance of the proposed method indicate that Swin Transformer model is an efficient tool for Non-Destructive Inspection of jet engine turbine and compressor blade surface which can be integrated into automated maintenance systems for better reliability and minimized operational risks.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

High-precision non-destructive blade surface inspection via self learning transformer networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月24日

查看全文
  • Priyadharshini Kannusamy,
  • D. Gayathri,
  • S. Mirdula,
  • Ramasubramanian Bhoopalan,
  • D. Manikandan,
  • Krishnaraj Ramaswamy

摘要

Accurate identification of defects in Jet Engine Turbine and compressor blade surface plays a vital role in ensuring engine efficiency, safety and secure life. Traditional detection techniques were manual and consumed more time, error prone and need of specialized experts. To address these challenges, this work proposes an automated defect detection system using Swin Transformer Based Deep learning model. High resolution blade surface images were captured, pre-processed and augmented to improve robustness with varying surface condition. The novelty of the proposed work is that it adapts the Swin Transformer architecture to the specific challenges of turbine and compressor blade surfaces and it can detect micro-scale defects with high fidelity, by detecting both local and global features. Experimental results demonstrated that the proposed Swin transformer model produces high detection performance compared to the conventional CNN model with an accuracy of 98.4%, precision of 97.9%, recall of 98.7%, F1-score of 98.3% and mean Average Precision (mAP) of 97.6% on a dataset consisting of 172 high-resolution turbine and compressor blade images. The performance of the proposed method indicate that Swin Transformer model is an efficient tool for Non-Destructive Inspection of jet engine turbine and compressor blade surface which can be integrated into automated maintenance systems for better reliability and minimized operational risks.